Put AI agents to work safely, with the evidence written down
A tool-agnostic deployment playbook for operations leaders in teams and small to mid-size companies: from choosing the first job to a signed pilot memo, with every record you need along the way.
- A 21-sheet workbook with a scored maturity self-assessment, registers, test cases, pilot monitoring and a dashboard, plus a business case calculator that counts review time and errors
- The autonomy ladder, a permission map and checkpoint records that set exactly what each agent action may do, with limits enforced in your systems
- Five fillable forms and seven Word documents, including the agent brief, a policy and incident runbook starter, the leadership sign-off memo and a message library
- A 45-minute team briefing with a 14-question LMS-ready knowledge check, and a dated sheet on the rules in the EU, UK, US, Canada, Australia, New Zealand and India
The AI Agent Deployment Playbook · Bodhih Pro Kits · 311 by Bodhih Training, trainers to 2,000+ companies since 2008.
How do you deploy AI agents safely in a small company?
Score candidate jobs on value, risk, data sensitivity and reversibility; give each action a rung on the autonomy ladder and a limit; use dedicated least-privilege accounts; put human approval before hard-to-undo actions; test on your own cases; pilot with weekly numbers; keep a tested kill switch. The AI Agent Deployment Playbook from Bodhih Training gives you every file.
Want the full free guide first? Read How to Deploy AI Agents Safely in a Small Business.
You already feel the problem. Here is what it is costing you.
The vendor says it can act on its own
Your help desk or finance tool now offers an agent that refunds, replies or pays. Nobody has decided how much it should do alone, or who answers when it is wrong.
It can reach far more than it needs
Agents are often connected with a person's login and broad connectors. One confused instruction and it can do things nobody intended.
The checks become clicks
Human approval sounds safe until a reviewer has 200 items a day. You need checkpoints people can actually keep up with.
Nobody can say whether it pays
Business cases count the licence and the hours saved, and forget review time and the agent's own mistakes.
There is no plan for the bad day
When an agent says something wrong or sends data to the wrong person, the first hour decides the damage. Most teams have no kill switch they have tried.
Picture the next few weeks with the kit open beside you.
Imagine the Friday of your first pilot week. You open the workbook's Dashboard and the numbers are there: items handled, the share approved without edits, escalations, errors and cost, each with a traffic light and a number. You know which actions sit at which rung, and why.
As you read the log sample, you notice something: when a draft is wrong, it is rejected with a reason, and that reason is already a new test case. The approval queue is a size people can manage. The agent's account can do only what the permission map says.
When something does go wrong, and in a real pilot it will, someone pulls a kill switch they have already tried, the fallback starts, and the incident report walks them through the first hour. The fix goes into the system, not just the wording.
Eight weeks later you write one page: what was tested, the results against the targets, the incidents, the residual risks and your recommendation. A leader signs it. Open the Start-here map tonight and give it one hour; by the end of it you will have a scored self-assessment and your first list of candidate jobs.
23 files that do the work for you
The e-book shows you how. The tools make it happen. Here is exactly what you download.
The AI Agent Deployment Playbook (e-book, PDF)
The guide: methods, worked examples, scripts and exercises
Start here: the AI agent deployment map
The job in ten steps with the file for each, your first hour, your first week, the 30/60/90-day plan and the 'Brand it in five minutes' box.
AI Agent Deployment Workbook
21 sheets: Settings, a scored maturity self-assessment with a ranked gap-to-action plan, use-case scoring, agent register, permission map, test cases, weekly pilot monitoring, costs, incidents, decision log, evidence register, vendor scoring, RACI, roadmap, jurisdiction crosswalk, official sources, What changed log, UK and US glossary, and a dashboard with four charts and traffic lights.
Agent Business Case Calculator
7 sheets: settings, cost today against cost with the agent per use case (including review time and error costs), payback and 12-month net, risk exposure ranges, cautious, expected and strong scenarios, a 12-month view and a dashboard with three charts.
Agent use-case intake and scoring card
Two-page fillable card for each candidate job: what the job is, who does it today, the five scores, the verdict and starting rung, the business owner and the decision. Copy the scores into the Use cases sheet.
Autonomy, permissions and checkpoint record
Three-page fillable record signed by the business owner for each agent: the rung and limit for every action, promotion and demotion rules, the permission map summary, the approval checkpoint design, escalation triggers and the weekly sample check.
Pilot plan, test and go-live record
Four-page fillable record: the pilot plan with scope, duration, metrics, targets and go/no-go criteria; the test set summary and pass bar; the plain-words red-team; and the go-live and end-of-pilot decisions.
Agent vendor due-diligence questionnaire
Four-page fillable questionnaire to send to each shortlisted vendor: security, data handling, controls, logs, integration, cost and exit, plus contract points for your lawyer and a scoring box for the Vendors sheet.
Agent incident and rollback report
Three-page fillable report: the first-hour runbook checklist, the incident record with severity and scope, the personal-data assessment, the rollback log and the review with root cause, system fix and new test.
Agent brief and instruction template
The agent's job description in eight sections (purpose, scope, approved sources, actions and limits, refusals, escalation, tone, disclosure), a paste-ready instruction block and a version-controlled change log.
Agent use policy, procedure and runbook starter
Part A a one-page agent use policy with 12 rules and roles; Part B the eight-step deployment procedure with records and sign-offs; Part C the incident, kill-switch and rollback runbook with severity levels, first-hour steps and contacts; Part D staff quick rules.
Leadership sign-off memo template
A one-page decision memo for go-live, rung changes, scale-up or stop: what was tested, results against targets, incidents, residual risks, recommendation, conditions and signature, with a filled Larchway example.
Agent deployment message library
34 ready-to-edit emails, chat messages and notices: announcing the pilot, answering 'will it replace us?', rung changes, customer disclosure and hand-off, correcting wrong information, incident and data notices, supplier bank-detail checks, vendor questions, leadership updates and outside enquiries.
Worked example pack: Larchway Home & Kitchen
Every main record filled in for one composite company: self-assessment, use-case scores, autonomy and permission records, brief extract, test results and go-live, 8 weeks of pilot numbers for both agents, the business case, incidents and the sign-off.
AI prompt pack for agent deployment
35 tool-agnostic prompts grouped by step (finding jobs, autonomy, permissions, checkpoints, briefs, testing, vendors, pilot, cost, incidents, people, governance), each with what to paste in and what to check, plus six human-review rules.
Team briefing session plan and answer key
A 45-minute briefing plan with timings, facilitator notes, a printable 14-question knowledge check, the answer key with explanations and an attendance record.
Autonomy ladder decision card and mistakes to avoid
The five rungs at a glance, 'if this, do that' decision rules for rungs, checkpoints and pausing, and the 12 mistakes teams make most with agents.
Agent failure modes and test card
20 common ways agents fail, the test that finds each one and the fix, with the seven test types and the pass bar.
Numbers that matter: agent KPI card
The 13 numbers an agent deployment runs on: plain meaning, exact formula, where each lives in the kit and what to do when it moves.
Agent rules by country (correct at November 2026)
The dated sheet: the EU AI Act timetable after the 2026 Digital Omnibus, and the main positions for the UK, US, Canada, Australia, New Zealand and India, each with the official source to check.
Knowledge check for the team briefing (LMS-ready CSV)
14 scenario-based multiple-choice questions with explanations. Imports into Bodhih LMS or any LMS that accepts CSV; the printable answer key is in the briefing session plan.
90-day agent deployment plan calendar
28 events from the first self-assessment to the sign-off meeting, with weekly pilot reviews and the first quarterly review. Import into Google Calendar, Outlook or Apple Calendar; starts on a Monday you can change.
Agent deployment task board (import file)
27 tasks with phase, owner, week and the kit file to open. Imports into Trello, Asana, Notion, Microsoft Planner, Jira or a spreadsheet.
Chapter by chapter
Each chapter opens with a real situation, gives you a method and ends with something to do today, using the tools in your kit.
- Format
- PDF e-book + XLSX + fillable PDF + DOCX + printable PDF + ICS + CSV
- Files
- 23 files plus a one-click ZIP
- E-book
- 73 pages
- Works with
- Excel, Google Sheets, Apple Numbers, LibreOffice, Word, Google Docs, Acrobat Reader, any LMS that imports CSV, Google Calendar, Outlook, Trello, Asana, Notion, Planner, Jira
- Best for
- Teams and companies of 10 to 500 people deploying vendor or platform AI agents
Find the Right First Jobs
Score every candidate job on value, feasibility, risk, data sensitivity and reversibility, and pick the first two with named owners.
The Autonomy Ladder
The five rungs from suggest to act alone, a rung and limit for every action, and written rules for moving up and down.
Permissions and Least Privilege
Dedicated agent accounts, the permission map, narrow tools and limits enforced in the systems themselves.
Human Checkpoints That Actually Work
Approval checkpoints people can actually keep up with, escalation triggers and the weekly check on the checkers.
Write the Agent's Brief
The eight-section agent brief, approved sources, keeping instructions apart from data, and version control.
Test Before You Trust
Test sets from your own work across seven types, a pass bar set in advance and a plain-words red-team.
Choose Vendors and Platforms
The same questions to every vendor on security, data, controls, logs, integration, cost and exit, weighted scoring and contract points.
Run the Pilot, Watch the Numbers
A time-boxed pilot with targets, weekly numbers from the logs, and an honest business case that counts review time and errors.
When It Goes Wrong
A tested kill switch, fallback and rollback, the first hour, personal-data assessment and fixing the system rather than the wording.
Bring People With You and Keep the Record
Briefing and involving staff, the agent register, decision log, evidence register, RACI, the crosswalk and the signed memo.
Not just reading. Tools you will actually open again.
Start here
- 00-Start-Here-AI-Agent-Deployment-Map.pdf: the job in ten steps with the file for each, your first hour, your first week, the 30/60/90-day plan and the 'Brand it in five minutes' box
Workbooks
- AI-Agent-Deployment-Workbook.xlsx: 21 sheets including Settings, a 20-practice maturity self-assessment with a ranked gap-to-action plan, use-case scoring, agent register, permission map, test cases, weekly pilot monitoring, costs, incidents, decision log, evidence register, vendor scoring, RACI, 13-week roadmap, jurisdiction crosswalk, official sources, What changed log, UK and US glossary and a dashboard with four charts and traffic lights
- Agent-Business-Case-Calculator.xlsx: 7 sheets: cost today against cost with the agent (including review time and error costs), payback, 12-month net, risk exposure ranges, three scenarios and a dashboard
Fillable forms
- Agent-Use-Case-Intake-and-Scoring-Card.pdf: the job, the five scores, verdict, starting rung and decision
- Autonomy-Permissions-and-Checkpoint-Record.pdf: rungs and limits per action, promotion and demotion rules, permissions, checkpoints and escalation, signed by the business owner
- Pilot-Plan-Test-and-Go-Live-Record.pdf: pilot plan and targets, test set and pass bar, plain-words red-team, go-live and end-of-pilot decisions
- Agent-Vendor-Due-Diligence-Questionnaire.pdf: security, data, controls, logs, integration, cost and exit questions, contract points and scores
- Agent-Incident-and-Rollback-Report.pdf: the first-hour runbook, incident record, personal-data assessment, rollback log and review
Word templates
- Agent-Brief-and-Instruction-Template.docx: the eight-section brief, a paste-ready instruction block and a change log
- Agent-Use-Policy-Procedure-and-Runbook-Starter.docx: a one-page policy with 12 rules, the eight-step procedure, the incident, kill-switch and rollback runbook and staff quick rules
- Leadership-Sign-Off-Memo-Template.docx: the one-page decision memo with a filled example
- Agent-Deployment-Message-Library.docx: 34 emails, chat messages and notices for staff, customers, suppliers, vendors and leaders
- Worked-Example-Pack-Larchway.docx: every main record filled in for the composite company in the book
- AI-Prompt-Pack-for-Agent-Deployment.docx: 35 tool-agnostic prompts with what to paste in, what to check and six human-review rules
Team training
- Team-Briefing-Session-Plan-and-Answer-Key.docx: a 45-minute session plan, a printable 14-question knowledge check, the answer key and an attendance record
- Agent-Knowledge-Check-LMS.csv: the 14 scenario questions, ready to import into Bodhih LMS or any LMS that accepts CSV
Cheat sheets
- Autonomy-Ladder-Decision-Card-and-Mistakes.pdf: the ladder at a glance, 'if this, do that' rules and 12 mistakes to avoid
- Agent-Failure-Modes-and-Test-Card.pdf: 20 failure modes with the test and the fix for each
- Numbers-That-Matter-Agent-KPI-Card.pdf: 13 numbers with formulas, where they live and what to do when they move
- Agent-Rules-by-Country-November-2026.pdf: the dated sheet with the EU AI Act timetable after the 2026 Digital Omnibus and the UK, US, Canada, Australia, New Zealand and India positions, with official sources
Calendar and import
- 90-Day-Agent-Deployment-Plan.ics: 28 events from the first self-assessment to sign-off and the first quarterly review
- Agent-Deployment-Task-Board.csv: 27 tasks for Trello, Asana, Notion, Planner or Jira
E-book
- The AI Agent Deployment Playbook: about 10,500 words in ten chapters with 24 figures, following one composite company's pilot from first score to signed memo
When you have worked through it, you will be able to…
Every outcome is practised with a worked example and a tool, not just described.
- Score candidate jobs and choose the first two for an agent pilot, with named owners
- Set a rung on the autonomy ladder and a limit for every agent action, with written rules for moving up and down
- Build a permission map and dedicated least-privilege accounts with limits set in your systems
- Design approval checkpoints and escalation triggers people can keep up with
- Write a version-controlled agent brief and test it on your own cases, including prompt injection
- Run a time-boxed pilot with weekly numbers and an honest business case
- Contain an agent incident in the first hour with a tested kill switch and rollback steps
- Brief your team, keep the register and evidence, and get a one-page memo signed
Know your starting point. Prove your progress.
People who measure before and after improve faster and can show it. AssessAll, our sister assessment platform, has AI-graded assessments that pair with this kit. Take one now, work through the kit, then re-take it.
AI Agent Oversight and Delegation Assessment for Knowledge Work
Take the assessment →AssessAll assessmentSigning It Off: AI Governance Judgment
Take the assessment →AssessAll assessmentTask Specification and Prompt Writing Work Sample for Knowledge Workers Using Generative Tools
Take the assessment →JobularyIf-then plans that make habits stick (Jobulary)
Make it a growth plan →A taste of Chapter 2
Priya split the customer service agent into actions. Looking up an order and replying about its status went to Act and report from week 3, after two weeks at Draft. Refunds up to C$75 went to Act with approval: the agent prepared the refund and a short reason, and one of Marcus's team approved it in the help desk with one click. Refunds above C$75 stayed with people. The invoice agent sat at Act with approval for everything: it matched invoices to purchase orders and goods received, prepared the payment batch and waited for Elena. The promotion rule was written in one line: "Move an action up one rung after four consecutive weeks with at least 85% of drafts approved unchanged and no Medium or higher incident." By week 8 the refund drafts had reached 86% approved unchanged, but the four-week clock had restarted after an incident in week 7, so they stayed where they were. The rule did its job: nobody had to argue.
The honest comparison
| Free templates and typical advice | This kit | |
|---|---|---|
| Choosing the first job | A list of AI ideas | Five-factor scoring with verdict rules, starting rungs and named owners |
| How much the agent may do | 'Keep a human in the loop' | A five-rung ladder per action, limits, and written promotion and demotion rules |
| Permissions | Not covered | A permission map, dedicated accounts and limits enforced in the systems |
| Testing | Try a few questions in the demo | 40 to 80 cases across seven types, a pass bar set in advance and a plain-words red-team |
| Cost and return | Licence price against hours saved | A calculator with review time, error costs, risk ranges and a cautious scenario |
| Against hiring a consultant or buying governance software | A consultant brings judgement and hours; software brings dashboards but not your decisions | The method, records and templates to run it yourself, with clear points where a lawyer or specialist should step in |
Made for you if…
- Chief operating officers, operations managers and general managers in companies of 10 to 500 people
- Customer service, finance and back-office leads whose software now offers 'agents'
- Founders who have been asked 'can we switch this on?'
- IT, security and privacy leads asked to approve an agent pilot
- Consultants and fractional COOs who set up agents for clients (consultancy licence via solutions@bodhih.com)
- Teams in the US, Canada, the UK, the EU, Australia and New Zealand, and other English-speaking markets
Probably not for you if…
- Not legal advice, an audit or a certification, and not a guarantee of compliance with any law
- Not for providers who build high-risk AI systems or general-purpose AI models for others
- Not a course in building AI models or writing code
- Not a full AI governance programme: The AI Use Policy and Governance Kit covers the wider policy, inventory and committee
Built by trainers, tested in real rooms.
Bodhih Training has designed and delivered corporate programmes since 2008. Bodhih Pro Kits package what works in those workshops into a guide plus the exact tools we use, so you can apply it the same day without a facilitator in the room.
From here to using it in three steps
- STEP 1
Tap Get the kit
Name, email and country. UPI, card or netbanking in India; international cards everywhere else. No account to create.
- STEP 2
Download instantly
Your download page opens at once, with every file and a one-click ZIP. A permanent link lands in your inbox too.
- STEP 3
Open the first tool
Start with the quick-start chapter, then the main spreadsheet or planner. Most buyers use something from the kit the same day.
Your first agent, deployed with the evidence written down
Open the Start-here map and give it one hour. Whether you begin tonight with the self-assessment or on Monday with the Use cases sheet, you will know which job to start with, how much the agent may do and how you will know it is working.
Personal licence for the buyer's own use, including use within your own company and with its staff. Please don't share or resell the files. Team, company, multi-site and consultancy licences: solutions@bodhih.com.
What buyers say
No ratings yet. Only people who have bought this kit can rate it - every buyer gets a one-click rating link a couple of days after their purchase.
The Bodhih family, when you are ready for more
The EU AI Act Deployer Compliance Kit
Get an organisation that uses AI ready for the EU AI Act as a deployer.
The AI Literacy Programme Kit
Design, run and evidence a workforce AI literacy programme.
AssessAll
250+ AI-graded assessments to measure skills and earn verified credentials.
Jobulary
Individual development plans that start from evidence and stay yours.
Bodhih Training
Run this topic as a live workshop for your team, in person or online.
Pewple
Book a vetted coach for career, leadership or communication. Pay per session.
Bodhih.org pathways
Certification pathways in Applied AI, Workplace English and management.
The AI Agent Deployment Playbook: FAQs
Read the free guideAlready bought? Resend my linkWhat exactly is in the kit?
An e-book, the 21-sheet AI Agent Deployment Workbook and a 7-sheet business case calculator (Excel, Google Sheets or Numbers, no macros), five fillable PDF forms (16 pages), seven Word documents including the policy and runbook starter, the sign-off memo, 34 messages and 35 prompts, four cheat sheets, a 45-minute team briefing with a 14-question LMS-ready knowledge check, a 90-day calendar and a task board.
What is the difference between an AI agent and a chatbot?
A chatbot drafts text that a person then uses. An agent takes actions in other systems: it looks up records, sends messages, issues refunds or prepares payments. Because a wrong action costs more than a wrong draft, agents need autonomy levels, permissions, checkpoints, testing and a kill switch, which is what this kit covers.
Do I need an AI agent policy if we already have an AI use policy?
Usually yes, or at least an agent section. A general AI use policy covers how staff use AI tools; agents act on their own and need rules on owners, autonomy levels, permissions, approval, testing, logs and incidents. The kit includes a one-page agent policy, procedure and runbook starter to adapt with your adviser.
How long does it take to deploy an AI agent safely?
The kit's plan runs 13 weeks: two weeks to score and select, two to design and test, an eight-week pilot and a sign-off week. Small, low-risk jobs can move faster; jobs that touch money or personal data deserve the full pilot.
Can I do this without a consultant?
Many small and mid-size teams can run the method themselves with the workbook, forms and templates. You will still want a lawyer for vendor contracts and an adviser for any legal question in your country, and you may want specialist help for complex integrations or high-risk uses.
Is this required by law in the UK, US, EU, Canada or Australia?
No single law requires this exact method, and the kit is not legal advice. Data protection, consumer protection and, in the EU, AI Act duties can apply to how you use agents, and they differ by country. The dated sheet and the Crosswalk sheet point to the official sources to check with your adviser.
Does it work in my country?
The method is the same everywhere. The dated sheet covers the EU, UK, US, Canada, Australia, New Zealand and India with official sources, correct at November 2026; laws change, so check the official page and your adviser locally. Currency, hourly costs and thresholds are all on the Settings sheets.
Can I use this with my team?
Yes. The personal licence covers use within your own company, including sharing the forms, the briefing and the knowledge check with your staff. Company-wide, multi-site and consultancy licences are available from solutions@bodhih.com.
Does it recommend a particular AI vendor or platform?
No. It is tool-agnostic: it gives you the questions, the scoring and the tests to choose and control any vendor's agent, including help desk add-ons, back-office tools and agent platforms.
What does it cost to run an AI agent pilot?
It depends on the vendor's pricing, your volumes and the staff time spent reviewing. The calculator counts platform and usage fees, set-up, review time and the cost of the agent's own errors, so you can see your own payback with your own estimates rather than a vendor's.
How do I measure my team's readiness to oversee agents?
Start with the kit's self-assessment and the knowledge check. If you want an independent measure of individual skill, AssessAll offers the AI Agent Oversight and Delegation Assessment for Knowledge Work, Signing It Off: AI Governance Judgment, and a Task Specification and Prompt Writing Work Sample.
How do I turn this into a development plan for the people involved?
Use the gap-to-action plan for the organisation and Jobulary for individuals: its guide to if-then plans helps reviewers and owners build habits such as 'if a draft looks wrong, then I reject it with a reason' that stick after the briefing.
How much does The AI Agent Deployment Playbook cost?
₹1,499 including 18% GST for buyers in India, or $39 for buyers outside India. One-time payment, no subscription, with an automatic invoice.
How do I get the files after paying?
Instantly. Your download page opens the moment payment succeeds, and a permanent download link is emailed to you. Download each file separately or everything in one ZIP.
What software do I need?
Any of the usual tools: Excel or Google Sheets for spreadsheets, Word or Google Docs for templates, Adobe Acrobat Reader, Preview, Xodo or GoodNotes for the fillable PDFs, and Google, Outlook or Apple Calendar for the calendar plan.
Can I get a refund?
Because it is a digital download, kits are not refundable once downloaded. We always refund a duplicate charge or a failed delivery, and if a file is corrupted and we cannot fix it within 3 working days. Full terms: bodhih.org/refunds.
Can my company buy this for a team?
Yes. Team and company licences, and workshops on this topic run by Bodhih Training, are available. Write to solutions@bodhih.com or call +91 99000 11601.
